Analyzing factors influencing defect-based conditions for sewer pipes using Bayesian networks

Abstract

The failure and condition of sewer pipes are usually influenced by various uncertainties and factors. Therefore, it is vital to evaluate the defect-based condition of sewer pipes for maintenance and failure prevention. In contrast to relying on the existing domain knowledge, this paper proposes a data-driven Bayesian network (BN) model to analyze the impacts of different influence factors (IFs) on the sewer pipes’ defect-based condition based on the developed database about the Hong Kong sewer network. Specifically, fourteen IFs are collected to develop a multi-source integrated database that incorporates pipe physical, environment, and climate-related factors. Then, the structure of the BN model is learned by Bayesian searching algorithm, and the reliability of the proposed model is evaluated using sensitivity analysis methods. Moreover, special scenarios are assumed to explore possible configurations of IFs. The results reveal that age, diameter, population and soil type are the top four IFs affecting the condition of sewer pipes, among which pipes with some characteristics need to be closely monitored, such as pipe age greater than 50 years, diameter less than 200 mm, location with a population density less than 15,000, and location in fill or granitic rocks. This paper improves sewer pipe management by integrating a comprehensive database and enabling reliable pipe condition inferences. The insights provide practical suggestions for the sewer pipe layout, risk analysis, and maintenance strategy formulation. It is a precious tool for authorities to enhance the safety and efficiency of sewer system management.